minima

Why we included this project

Retrieval over your own documents usually means choosing between a hosted service and running an index yourself. Minima is the second option with the plumbing already wired: a containerized stack that indexes local PDFs, Word, Excel, Markdown, CSV and text files with sentence-transformers embeddings in Qdrant and answers questions over them. You can run the whole thing on your own hardware with Ollama, including a reranker, or connect any OpenAI-compatible endpoint such as vLLM, TGI or LocalAI and skip the reranking step. If your team already lives in ChatGPT or the Claude app, the same index can be reached through a custom GPT or an MCP server, so people query private documents from an interface they already know. Setup is driven by a compose file and an .env template, a realistic afternoon of work for a small team that wants control over where every model runs.